Defect Inspection Feature Totalization for Sensitivity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect inspection methods face challenges in maintaining inspection sensitivity when the number of true information is insufficient, particularly due to the integration of multiple detection systems and segmentation of inspection regions, leading to difficulties in distinguishing true information from false information, especially when detecting fine defects near the detection limit.
Innovation Solution
A defect inspection device and method that includes an image generation unit, feature amount calculation, storage, and totalization units to determine the expected number of defects, allowing for the adjustment of extraction conditions without compromising sensitivity, even when true information is scarce, by integrating feature amounts from multiple detectors and regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If multiple detection systems are integrated to detect various defects, then the comprehensiveness of defect detection is improved, but the number of true information becomes insufficient
Solution Approach 1:
The patent combines image information from multiple detection systems (upper detection system and oblique detection system) to perform integrated defect inspection. By merging the detection results and feature amounts from both systems, the patent achieves comprehensive defect detection while efficiently utilizing the limited true information across all detection systems rather than requiring sufficient true information from each individual system.
2Measurement precision
If inspection region is segmented to facilitate discrimination between true information and false information, then the discrimination accuracy is improved, but the required number of true information increases
Solution Approach 1:
The patent divides the inspection region into multiple regions (first inspection region and second inspection region) and performs separate defect extraction for each region using region-specific extraction conditions. This segmentation allows optimization of detection parameters for different areas, improving discrimination accuracy between true and false information. The patent sets different extraction conditions for different regions, enabling effective use of limited true information from each segmented region rather than requiring a large total number of true information across the entire inspection area.
3Measurement precision
If inspection sensitivity is increased to detect fine defects near detection limit, then the defect detection capability is improved, but the ability to discriminate true information from false information deteriorates
Solution Approach 1:
The patent changes extraction conditions (parameters) according to the inspection region and defect type. By adjusting extraction conditions for different regions and using multiple detection systems with different detection characteristics, the patent achieves high sensitivity detection of fine defects while maintaining discrimination capability. The system extracts feature amounts under multiple extraction conditions and synthesizes the results, enabling both high sensitivity and effective discrimination between true and false information.
4Object-generated harmful factors
If parameters are updated by repeatedly teaching with non-defective pixels until no non-defect is detected as defect candidate, then the false information reduction is improved, but the inspection sensitivity deteriorates
Solution Approach 1:
Instead of repeatedly updating parameters until perfect discrimination is achieved (excessive action), the patent performs defect extraction under multiple extraction conditions and synthesizes the results. This partial action approach prevents over-optimization that would reduce sensitivity. By using multiple extraction conditions and combining results, the patent reduces false information while maintaining inspection sensitivity, avoiding the need for excessive parameter updates that would eliminate true defects along with false information.
Data Source
AI summary
A defect class that can identify true information or false information regarding a defect of an inspection target is received, it is determined whether or not the number of true information and the number of false information are insufficient to set a defect extraction parameter of a predetermined region, a feature amount of the defect is extracted in a region other than the predetermined region when it is determined that the number of true information and the number of false information are insufficient, the feature amounts are totalized by adding the feature amount of the defect extracted in a region other than the predetermined region to the feature amount of the defect extracted in the predetermined region, and the totalized feature amount of the defect is displayed for adjustment of the defect extraction parameters.